Precise temperature measurement during the growth of deposited layers on a semi-conductor wafer is critical to the ultimate quality of the finished, coated wafer and in turn to the performance of the opto-electronic devices constructed on the wafer. Variations in substrate temperature, including intra-wafer variations in temperature ultimately affect quality and composition of the layers of material deposited. During the deposition process, the substrate wafer is normally heated from behind and rotated about a center axis. Typically, a resistance heater positioned in proximity to the wafer provides the heat source for elevating the temperature of the wafer to a pre-determined value. Precise control of the temperature associated with the process is most desirable, and is best achieved through precise and real-time monitoring of the substrate temperature.
An example application illustrating the necessity of precise temperature control is the formation of semiconductor nanostructures. Semiconductor nanostructures are becoming increasingly important for applications such as “quantum dot” detectors, which require the self-assembled growth of an array of very uniform sizes of nano-crystallites. This can only be accomplished in a very narrow window of temperature. Temperature uncertainties can result in spreading of the size distribution of the quantum dots, which is detrimental to the efficiency of the detector.
The growth of uniform quantum dots is an example of a thermally activated process in which the diffusion rates are exponential in temperature. Therefore, it is important to be able to measure, and have precise control over, the substrate temperature when growth or processing is performed.
Numerous methods have been disclosed for monitoring these temperatures. One simple, but largely ineffective approach has been the use of conventional thermocouples placed in proximity to, or in direct contact with the substrate during the thin film growth operation. This methodology is deficient in many respects, most notably, the slow response of typical thermocouples, the tendency of thermocouples (as well as other objects within the deposition chamber) to become coated with the same material being deposited on the semi-conductor wafer, thereby effecting the accuracy of the thermocouple, as well as the spot thermal distortion of the surface of the semiconductor wafer resulting from physical contact between the thermocouple and the substrate. In any event, the use of thermocouples near or in contact with the substrate is largely unacceptable during most processes because of the poor accuracy achieved.
Optical pyrometry methods have been developed to overcome these shortcomings. Optical pyrometry uses the emitted thermal radiation, often referred to as “black body radiation,” to measure the sample temperature. The principal difficulties with this method are that samples typically do not emit sufficient amounts of thermal radiation until they are above approximately 450° C., and semiconductor wafers are not true black body radiators. Furthermore, during deposition a semiconductor wafer has an emissivity that varies significantly both in time and with wavelength. Hence the use of pyrometric instruments is limited to high temperatures and the technique is known to be prone to measurement error.
In “A New Optical Temperature Measurement Technique for Semiconductor Substrates in Molecular Beam Epitaxy,” Weilmeier et al. describe a technique for measuring the diffuse reflectivity of a substrate having a textured back surface, and inferring the temperature of the semiconductor from the band gap characteristics of the reflected light. The technique is based on a simple principle of solid state physics, namely the practically linear dependence of the interband optical absorption (Urbach) edge on temperature.
Briefly, a sudden onset of strong absorption occurs when the photon energy, hv, exceeds the bandgap energy Eg. This is described by an absorption coefficient,α(hv)=αgexp[(hv−Eg)/E0],where αg is the optical absorption coefficient at the band gap energy. The absorption edge is characterized by Eg and another parameter, E0, which is the broadening of the edge resulting from the Fermi-Dirac statistical distribution (broadening ˜kBT at the moderate temperatures of interest here). The key quantity of interest, Eg, is given by the Einstein model in which the phonons are approximated to have a single characteristic energy, kB. The effect of phonon excitations (thermal vibrations) is to reduce the band gap according to:Eg(T)=Eg(0)−SgkBθE[exp(θE/T)−1]where Sg is a temperature independent coupling constant and θE is the Einstein temperature. In the case where θE>>T, which is well-obeyed for high modulus materials like Si and GaAs, one can approximate the temperature dependence of the band gap by the equation:Eg(T)=Eg(0)−SgkBT, showing that Eg is expected to decrease linearly with temperature T with a slope determined by Sg kB. This is well obeyed in practice and is the basis for the band edge thermometry.
Variations on this methodology are taught by Johnson et al., in U.S. Pat. No. 5,388,909, and U.S. Pat. No. 5,568,978. These references teach the utilization of the filtered output of a wide spectrum halogen lamp which is passed through a mechanical chopper, then passed through a lens, then through the window of high vacuum chamber in which the substrate is located, and in which the thin film deposition process is ongoing. Located within the chamber is a first mirror which directs the output of the source to the surface of the substrate. The substrate is being heated by a filament or a similar heater which raises the temperature of the substrate to the optimum level required for effective operation of the deposition process. A second mirror located within the chamber is positioned to reflect the non-specular (i.e., diffuse) light reflected from the back surface of the substrate, said reflection being directed to another window in the chamber and thence through a lens to a detection system comprising a spectrometer. The wavelengths of the elements of the non-specular reflection are utilized to determine the band gap corresponding to a particular temperature. Johnson et al. teach that the temperature is determined from the “knee” in the graph of the diffuse reflectance spectrum near the band gap.
While the prior art is in some ways effective, use of optical fiber bundles, intra chamber optics, mechanical light choppers and mechanically scanned spectrometers renders the methodology deficient in many respects. The detected signal suffers from temporal degradation of the optics within the deposition chamber. The mechanical components are overly susceptible to failure and the overall methodology of collecting the signal is simply too slow for real-time measurement and control applications in the industrial production environment. In addition, the described means of the prior art is subject to variations in accuracy dependent upon the fluctuation, over time, of the output of the halogen light source.
Specifically, the prior art relies on one or more optical elements within the deposition chamber to direct the incident light to the wafer and to collect the diffusely reflected light. The presence of optics within the deposition chamber is problematic, since the material being deposited during the coating process tends to coat all of the contents of the chamber, including the mirrors, lenses, etc. Over time the coatings build up and significantly reduce the collection efficiency of the optics and can lead to erroneous temperature measurement.
More importantly, the prior art relies on a mechanical light chopper and a mechanical scanning spectrometer for measurement of the light signal. Not only do the mechanical components fail frequently with extended use, but it is well known that gears in scanning spectrometers wear, resulting in continual shifts in the wavelength calibration. This leads to perpetually increasing errors in temperature measurement unless the instrument is recalibrated frequently, which is a very time consuming process. In addition, it is well known that scanning spectrometers are quite slow, requiring anywhere from 1-5 seconds to complete a single scan. In most deposition systems the semiconductor wafers are rotating, typically at 10-30 RPM. In this case, a temperature measurement that takes 1-5 seconds to complete is by default an average temperature and it is impossible to make any type of spatially resolved measurement. If the process chamber has many wafers rotating on a platter about a common axis, as is typical in a production deposition system, the slow response time of the prior art makes it impossible to monitor multiple wafers.
Furthermore, the prior art utilizes a quartz halogen light source with no consideration of any type of output stabilization or intensity control. Quartz halogen lamps are known to degrade rapidly over time leading to fluctuations in the lamp output that result in measurement variations and further system downtime for lamp replacement.
Basically, the many limitations of the prior art have limited the applications of diffuse reflectance or “band edge” thermometry in the commercial setting.